Approaches for identifying PM2.5 source types and source areas at a remote background site of South China in spring
Graphical abstract
Introduction
Atmospheric PM2.5 has been of particular interest these days due to the negative impacts on health (Lelieveld et al., 2015; Heft-Neal et al., 2018) and climate change (Zou et al., 2017). In numerous studies of PM2.5, source analysis is one of the research priorities and has been conducted over the past three decades (e.g., Thurston and Spengler, 1985; Maykut et al., 2003; Lai et al., 2019), due to its vital implications on control strategies (Cai et al., 2018). In the developed methods, receptor model is a widely used tool for PM2.5 source apportionment and has been originally proposed by Blifford and Meeker (1967). It is an analytical method using particulate components observed in a receptor site to inversely determine source contributions. The receptor model includes many types, such as principal component analysis (PCA), enrichment factors (EF), chemical mass balance (CMB), Unmix, and positive matrix factorization (PMF) (Zoller et al., 1974; Winchester and Nifong, 1971; Zheng et al., 2014). Among, the CMB, Unmix, and PMF models are officially recommended by the US Environmental Protection Agency (USPEA) (www.epa.gov/ttn/scram/receptorindex.htm). However, CMB doesn't permit a quantitative source apportionment of particle mass and needs to be combined with other methods to achieve the quantification. In order to quantify the source contribution, Unmix has been developed by Henry and Kim (1990). However, it remains problems in algorithms (Paatero and Tapper, 1993) and can't separate sources which have a similar contribution to particle mass. Then, PMF model overcomes these problems using a least-square method to obtain the source profiles and their contributions (Hopke, 1985; Paatero, 1997).
PMF usually computes some factors that cannot correspond to explicit source classes (Watson et al., 2008) and its convergence result is not clear, leading to the difficulty in result selection (https://www.epa.gov/air-research/models-tools-and-databases-air-research). Thus, chemical species as source markers have been incorporated into the PMF model to determine and optimize the source profiles of PM2.5 such as organics and radiocarbon (e.g., Choi et al., 2015 and Zong et al., 2016). In addition, PMF requires a large number of input data sets, typically larger than 100 samples as recommended by Hopke, to obtain a reliable solution. For few samples, therefore, the PMF operation has recently been combined with other receptor models to compare their results and examine its accuracy of predictions in PM2.5 source apportionment, such as PMF-PCA and PMF-CMB, etc. (e.g., Gibson et al., 2015, sample n = 45; Al-Naiema et al., 2018, n = 44). Among these models, PMF is considered to be the most robust model for PM2.5 source apportionment even in the case of a small sample size. In addition, a non-negative matrix factorization (NMF) model with a similar algorithm to PMF has also been operated independently to obtain PM2.5 source profiles of 70 samples (Shang et al., 2018). However, an inter-comparison regarding to the coupled NMF model (e.g., NMF-PMF) has not previously been performed with a small size of samples.
Moreover, most source apportionment studies of PM2.5 focused on the polluted regions such as the port city of Thessaloniki, Greece (Saraga et al., 2019), the industrial coastal city of Houston, USA (Al-Naiema et al., 2018), the metropolitan city of Beijing, China (Yu et al., 2019) and Seoul, South Korea (Heo et al., 2009), the semi-urban area of Malaysia (Khan et al., 2016), the biomass burning prevailed Indo-Gangetic plain (Singh et al., 2017), the remote background site in Duolun (continent), China (Wang et al., 2008) and at Gosan (island), Korea (Han et al., 2006), and others particularly continental East, South, and Southeast Asia. To our knowledge, however, there are currently no source apportionment studies of PM2.5 conducted at a remote background site in the most southern region of China, where biomass burning, as a significant source of PM2.5, is prevalent especially during spring (e.g., Chen et al., 2017; Li et al., 2017).
Source apportionment of PM2.5 therefore was conducted at a remote background site of Weizhou Island, Beihai, South China, which locates at the downwind pathways of air masses transported from Southeast Asia, continental South China, and Taiwan Strait during spring (Zhou et al., 2018). This is the first field study on source analysis of PM2.5 accomplished on the island and therefore filter sampling continued for a short period (~1 month) as a preliminary measurement. Consequently, NMF and PMF model were selected because both have been successfully applied to estimate the PM2.5 source types and their contributions on the continents of a small sample size as mentioned above. The reason that two receptor models were combined in this study is that there was no prior information about sources in this area and their combination was expected to produce a reliable result by comparison. Meanwhile, the model results were introduced into trajectory models, including potential source contribution function (PSCF) and concentration-weighted trajectory (CWT) model, to determine the potential source areas and their contribution to the PM2.5 source (Liu et al., 2017). Also, PCA method and chemical characteristics of PM2.5 were incorporated to assist interpretation of the source analysis. The work is important for the efficient policy-making to reduce PM2.5 in the most southern China since source apportionment in regional scale cannot be replaced by that extensively conducted in the polluted regions.
Section snippets
Sampling and measurements
Sampling station (21°02′N, 109°06′E) was built at 15 m above the ground in the most southwest of Weizhou Island, Beihai, Guangxi province, South China. As a remote site, it is often influenced by various continental outflow plumes (e.g., Southeast Asia, continental South China, Taiwan) under dominant monsoons particularly in spring (Zhou et al., 2018). It covers an area of 25 km2 but with a population of only ~16,000 and less industrial factories. Therefore, the site is taken as an ideal
Measurement overview
During the entire experiment period, the total concentration of measured compositions in PM2.5 ranged from 7.65 to 27.97 μg m−3 (average of 16.32 μg m−3), which were nearly same as those in a previous study (8.06 to 27.0 μg m−3, average of 16.2 μg m−3) conducted on the east coast of peninsular Malaysia in winter as both local and transported heavy pollution plumes (from Southeast and East Asia) influenced (Farren et al., 2019). All measurements comprised 37%–64% (average of 51%) of the total PM
Cluster analysis
Cluster analysis was conducted, on the one hand, to identify the transport pathways of total air masses. Four clusters were computed (SI. 2) and the relevant sample size for each cluster was shown in Table 1. Cluster 1 (45.2%), sourced from Southeast Asia. The air masses of Cluster 2 passed over South China Sea accounting for 9.7% of the total. Cluster 3 and Cluster 4 originated from Taiwan Strait (25.8%) and Pearl River Delta region (19.4%), respectively.
On the other hand, cluster analysis was
Source types of PM2.5
Prior to the source apportionment of PM2.5, the performance of the NMF and PMF model was assessed based on the comparison of their predictions. The comparison between observation and prediction for both models focused on several source markers (secondary ions, K+, Ca2+, Mg2+, Mn, Fe, Pb, and levoglucosan), which were used to identify source factors in following source apportionment by NMF. The assessment parameters included signal-to-noise ratio (S/N) and regression (prediction and observation)
Conclusions
For the first time, the source types and source areas of PM2.5 were analyzed using a NMF method coupling with several hybrid receptor models (PMF, PSCF, CWT, and cluster analysis) at a remote background site of South China during a short period of spring, 2015. By comparing with the PMF model, NMF model was found to be more robust to predict chemical species and thereby was more reliable to obtain the source types and source contributions. In the experiment period, four main sources were
Declaration of Competing Interest
The authors declare no conflict of interest.
Acknowledgments
This work was supported by the National Department Public Benefit Research Foundation (No. 201309016, No. 201109005), the National Research Program for Key Issues in Air Pollution Control (No. DQGG0304-05) and the Fundamental Research Funds for Central Public Welfare Scientific Research Institutes of China (No. 2016YSKY-025). Special thanks give to Professor Gangwoong Lee for providing the package of NMF model.
References (52)
- et al.
A factor analysis model of large scale pollution
Atmos. Environ.
(1967) - et al.
Impact of air pollution control policies on future PM2.5 concentrations and their source contributions in China
J. Environ. Manage.
(2018) - et al.
A review of biomass burning: emissions and impacts on air quality, health and climate in China
Sci. Total Environ.
(2017) - et al.
Molecular marker characterization and source appointment of particulate matter and its organic aerosols
Chemosphere
(2015) - et al.
Trends of ambient fine particles and major chemical components in the Pearl River Delta region: observation at a regional background site in fall and winter
Sci. Total Environ.
(2014) - et al.
Extension of self-modeling curve resolution to mixtures of more than three components: Part 1. Finding the basic feasible region
Chemom. Intell. Lab. Syst.
(1990) - et al.
Comparison of hybrid receptor models to locate PCB sources in Chicago
Atmos. Environ.
(2003) - et al.
Chemical composition and source apportionment of ambient, household, and personal exposures to PM2.5 in communities using biomass stoves in rural China
Sci. Total Environ.
(2019) - et al.
Characterization and source apportionment of PM2.5 based on error estimation from EPA PMF 5.0 model at a medium city in China
Environ. Pollut.
(2017) Least squares formulation of robust non-negative factor analysis
Chemom. Intell. Lab. Syst.
(1997)
Analysis of different modes of factor analysis as least squares fit problems
Chemom. Intell. Lab. Syst.
PM2.5 source apportionment for the port city of Thessaloniki, Greece
Sci. Total Environ.
Fine particulates over South Asia: review and meta-analysis of PM2.5 source apportionment through receptor model
Environ. Pollut.
Chemical characteristics of haze during summer and winter in Guangzhou
Atmos. Res.
A quantitative assessment of source contributions to inhalable particulate matter pollution in metropolitan Boston
Atmos. Environ.
Long-term monitoring and source apportionment of PM2.5/PM10 in Beijing, China
J. Environ. Sci.
Chemical characteristics of water-soluble organic compounds (WSOC) in PM2.5 in Beijing, China: 2011–2012
Atmos. Res.
Identification of source nature and seasonal variations of Arctic aerosol by the multilinear engine
Atmos. Environ.
Characteristics and oxidative potential of atmospheric PM2.5 in Beijing: source apportionment and seasonal variation
Sci. Total Environ.
Fine particles (PM2.5) at a CAWNET background site in Central China: chemical compositions, seasonal variations and regional pollution events
Atmos. Environ.
Characterization of humic-like substances in PM2.5 during biomass burning episodes on Weizhou Island, China
Atmos. Environ.
Source apportionment of fine particulate matter in Houston, Texas: insights to secondary organic aerosols
Atmos. Chem. Phys.
Levels and composition of ambient particulate matter at a mountainous rural site in northern Vietnam
Aerosol Air Qual. Res.
Chemical characterisation of water-soluble ions in atmospheric particulate matter on the east coast of Peninsular Malaysia
Atmos. Chem. Phys.
A comparison of four receptor models used to quantify the boreal wildfire smoke contribution to surface PM2.5 in Halifax, Nova Scotia during the BORTAS-B experiment
Atmos. Chem. Phys.
Size-resolved source apportionment of ambient particles by positive matrix factorization at Gosan background site in East Asia
Atmos. Chem. Phys.
Cited by (26)
Trade-off efficiency: Linking urban socioecological quality with land use efficiency from return on investment perspective
2023, Sustainable Cities and SocietyConcentration prediction and spatial origin analysis of criteria air pollutants in Shanghai
2023, Environmental PollutionAn improved nonnegative matrix factorization with the imputation method model for pollution source apportionment during rainstorm events
2023, Journal of Environmental ManagementCitation Excerpt :In addition, compared with other receptor models, NMF uses multiplication iteration rather than addition or subtraction iteration. This algorithm ensures the nonnegativity of the results with less data adjustment to improve responsive original characteristics and reflects practical meanings (Zhang et al., 2019). Additionally, NMF is easily improved through different implementation coding methods because of its publicity (Janecek and Tan, 2011; Sharma et al., 2020).
Spatial origin analysis on atmospheric bulk deposition of polycyclic aromatic hydrocarbons in Shanghai
2022, Environmental PollutionCitation Excerpt :In order to reduce the risk from PAH deposition, it is necessary to locate the geographic source areas of atmospheric PAHs and then propose effective abatement strategies. Many studies have used Lagrangian trajectory models (e.g., HYSPLIT) to track pathways of air parcels (Hosseini Dehshiri et al., 2022; Stein et al., 2015; Sun et al., 2017) and locate source areas of atmospheric pollutants (e.g., PM2.5 and SO2) via probabilistic models, e.g., concentration weighted trajectory (CWT) and potential source contribution function (PSCF) models (Polissar et al., 2001; Sheng et al., 2013; Zhang et al., 2019). But only a few studies conducted the spatial location of emission source areas for atmospheric deposition of PAHs, for example, by discussing the influence of monsoon or correlations with air-mass backward trajectories (Arellano et al., 2018; Hu et al., 2021), possibly due to the lack of high frequency data or a limited sample size (i.e., <100 samples per site) to support the probabilistic models (Hafner and Hites, 2003).
New insight into the urban PM2.5 pollution island effect enabled by the Gaussian surface fitting model: A case study in a mega urban agglomeration region of China
2022, International Journal of Applied Earth Observation and GeoinformationCitation Excerpt :As in previous PM2.5 pollution studies, the quantitative characterization of the UPI effect is mainly based on two types of data sources, PM2.5 concentration data recorded by ground-based monitoring stations and data derived from remote sensing platforms (Hoek et al., 2008; Xie et al., 2015). Ground-based data offer the advantage of providing continuous observations (i.e., high temporal resolution to enable hourly or even finer recording) (Zhang et al., 2019), and provide the necessary reference for interpreting PM2.5 concentrations based on remote sensing sources (Van Donkelaar et al., 2010, 2015). However, ground-based data are often constrained by the lack of monitoring stations (due to, e.g., insufficient financial support, suboptimal local conditions, etc.) to provide continuous spatial representations of regional pollution conditions, especially in developing countries or regions suffering from serious air pollution problems, including China (Zhang et al., 2020a).